Re: Leak objects?

[Richard Frith-Macdonald]

On 3 Apr 2011, at 11:59, Fred Kiefer wrote:

> On 30.03.2011 10:54, Fred Kiefer wrote:
>> When running GSTest and checking the memory usage with the build in
>> memory panel, I noticed that when the NSAimation test is running we seem
>> to be leaking GSMutableArray objects. I remember resolving an issue in
>> that test a long time ago and I was convinced that this test wont leak
>> objects. Looks like something changes since then.
>> Could somebody please check, whether changed behaviour in base leads to
>> this result?
>> I don't think this potential bug is a release stopper. But it would be
>> good to understand whether this is specific to my machine (64 bit
>> OpenSuse 11.4) or a general problem.
> 
> I think I found the reason for this leak, but I would prefer for Richard to 
> look into this before applying a patch.
> 
> I think it is the class  GSRunLoopThreadInfo (in NSThread.m) that leaks its 
> ivar performers. A simple RELEASE(performers) in the dealloc method should 
> fix this.

Great ... thanks very much.  That's well spotted.

> I also had a quick look into the fire method and don't quite understand why 
> we copy the array there. Wouldn't it be much fast to replace the ivar with a 
> new empty array and reuse the original array in the loop?
> But this isn't a bug fix and shouldn't go in now.

I agree that replacing would be faster ... but perhaps I was contemplating 
using imp caching ... for which we'd want to keep the same object to avoid 
having to re-cache later.  I suspect that optimisation either way is not really 
worthwhile, but we could look into it later.